A buyer describes what they need to a chatbot and gets three supplier names back. You are not one of them, though your product fits, your certifications are current and your lead times beat all three.
Getting your company recommended by AI means publishing the exact facts a chatbot checks during a live buyer conversation: dimensions, certifications, lead times, prices, and industry-specific use cases as machine-readable text on your website. Without those values on a page a fetcher can read, you drop off the shortlist even when your product fits every condition the buyer named. The gap is not your offer. It is what never made it onto a page a machine can quote.
In this article:
- What changed about the way buyers find suppliers?
- Why does an AI leave out a company that fits?
- Which facts does an AI check, and what happens when they are missing?
- Where else does the answer come from?
- How do you publish all of that without hiring a department?
- What is the prize at the end of this work?
- How we build this in Drupal for our clients
What changed about the way buyers find suppliers?
Buyer behaviour moved quickly here. In March 2026 G2 surveyed 1,076 B2B software buyers, and 51% of them said they now begin research with an AI chatbot more often than with Google. Eleven months earlier only 29% said that. Seventy-one percent use a chatbot somewhere in the process.
Broader research shows the same thing outside software. Forrester's Buyers' Journey Survey, 2025 asked nearly 18,000 business buyers and found 94% using AI somewhere in their buying process, up from 89% a year earlier. Twice as many buyers called generative AI or conversational search a more meaningful source of information than anything else, vendor websites and sales conversations included.
Take somebody buying a pump for a yacht.
A few years ago they typed "bilge pump" into Google, opened six tabs and compared what they found. When no page said whether it fits under the floor panel, they called a distributor and asked. Your website did not have to be complete. The phone call finished the job.
Now they open a chatbot and write a paragraph instead of a phrase: "The pump has to fit a 400 mm locker. It runs off the boat's 12 V circuit and cannot draw more than 15 A. It sits in salt water and goes fully submerged from time to time. The water it moves carries grit."
Then they keep asking, because a conversation invites it. How loud is it. Which one has spare parts in Europe. Which has a stainless housing instead of a coated one. Every answer narrows the field.
The reply is a list of companies that make a pump like that, rather than a page of links. Three names, maybe four, and the buyer picks from those.
Your pump could meet every one of those conditions and still be missing from that list, and nobody called to check. Getting recommended by AI at this stage is not about brand awareness. The reason has nothing to do with the pump.
Why does an AI leave out a company that fits?
A chatbot answers from two sources. Part of it is memory from training, which is old and approximate. The rest it fetches live, sending a bot to open your pages while the buyer waits, looking for whatever was just asked about.
Each condition then gets checked against a value it can point to on a page. Publish the value and you stay in the running. Leave it out and you drop off the list even though you met the requirement, because nothing on the page proved that you did.
"Contact us for details" reads like silence, and so do "flexible", "tailored" and "comprehensive".
For a chatbot, anything you do but never wrote down is something you do not do. The certification you hold but never listed. A material you supply on request, or a service you have delivered for eleven years without ever describing it on a page, because everyone inside the company already knows about it.
The same G2 survey measured what that costs. AI chatbots came out as the strongest influence on which vendors reach a shortlist, ahead of review sites, analyst firms and vendor websites, and 69% of those buyers ended up choosing a different vendor than the one they had planned on. G2's own conclusion is blunt: once a chatbot leaves a vendor out, the buyer may never learn that the vendor exists.
You will not see that in any report. There is no impression, because you were never shown, and no click or bounce, because nobody arrived. Your traffic looks like last month's while somebody picks a competitor in a conversation you never see.
Read also: component mindset: teaching clients to think in components and don't rebuild, evolve: a phased CMS modernization framework.
Which facts does an AI check, and what happens when they are missing?
Content was always king. Now it has to be precise if you want your company recommended by AI in a buyer shortlist.
A chatbot recommends what it can verify, which turns the text on your website into your whole case. Nothing else is available for a machine to read.
Not any text, though. It has to be current, accurate and specific: numbers, dimensions, capacities, prices, lead times, limits, and honest descriptions of who uses the thing and what for. A chatbot is looking for facts: a dimension, a load, a tolerance, a range, a throughput, an ingress rating, a certification number, a lead time, a price, a count of supported languages.
Marketing language returns nothing here. It never worked very well on buyers either, but it used to be harmless, because a salesperson came along afterwards and filled the gap. Now nobody comes along, and a claim nobody can check may as well not be on the page. Descriptions of facts get nowhere. A "compact housing" is not a dimension, and a "wide range of sizes" is not a range. Run the yacht pump's conditions against a page written in adjectives and every condition fails, though the pump itself would have passed all five.
The chatbot will not call you, email you, fill in your contact form or wait for a datasheet to arrive. It reads what is published and answers from that.
So when a buyer asks whether the pump fits a 400 mm locker and no dimensions appear anywhere on your site, the buyer does not hear "ask the supplier". The buyer hears that the information is not available, and the competitor whose page carries the number gets recommended.
The sections below cover the gaps we see most often on B2B and manufacturing sites.
"Used across many industries" is probably the worst
Almost every company has that sentence somewhere. It tells a chatbot nothing, because nobody asks about many industries. They ask about their own.
A chatbot reading it cannot confirm that you serve food processing, so a buyer looking for suppliers with food processing experience will not see your name in the answer. Your product had nothing to do with that.
The way out is a list. Name every industry you serve, and give each one a page with the use cases that industry recognises, the constraints specific to it, the approvals it demands and a case from it. Ten industries means ten pages of facts.
The four things buyers ask about and websites refuse to publish but should
Before a buyer shortlists anyone, four questions get settled: what it costs, what it does and does not do, how it compares with the alternatives, and whether it has been done before for a company like theirs.
Most websites answer none of them. And a machine has no way to tell a company that keeps its prices private from a company that cannot do the job, so both end with your name missing from the answer.
Your website has to carry more than it used to
Coverage gets decided question by question, because that is how buyers ask. One page answering ten industries answers none of them.
That pushes the volume up. A website used to be a brochure, and its job was to get somebody interested enough to make contact; the detail followed later from a person. Every question that person used to field now needs a published answer.
Multiply that out: ten industries, forty products, six languages, plus the price, limits and comparison questions most sites avoid. Companies that get this far usually conclude that their website needs improving. It needs to get bigger and then stay accurate at that size, and staying accurate is the harder half.
Your website has to be readable and understandable for AI
Publishing the facts is half the work. They also have to arrive in a form a machine can read without guessing.
Every link in that chain can break quietly, since your own browser shows you a perfectly good page either way.
- Nothing blocks the way in. The robots file tells crawlers what they may read, and plenty of sites still run a copy from a staging server where everything was blocked on purpose. A firewall can do the same: it sorts unfamiliar traffic, an AI fetcher looks a lot like a scraper, so it gets a 403 instead of a page. That call belongs to whoever runs the network, and marketing usually never hears about it.
- The text sits in the HTML. If your pages assemble their content in the browser with JavaScript, a fetcher that does not run JavaScript receives a nearly empty file. You see a full page where it sees almost nothing.
- The values are text, not pictures. A dimension printed inside a datasheet image, a specification table that exists only as a scan, a price list living in an attached PDF: all of it reads as missing. Read also: why image-based content kills visibility, and how to fix it
- The markup says what each value means. Structured data labels a number as a price, a lead time or a capacity, instead of leaving a machine to work it out from the sentence around it. Markup alone will not make you quotable, but it removes ambiguity in the places where ambiguity costs you most.
- Nothing important hides behind a form or a login. Gate a fact and it stops existing for the shortlist.
- The same fact reads the same everywhere. One unit, one label, one value across the product page, the datasheet and the comparison table. Two numbers that contradict each other do more damage than a number you never published.
Most of this sits in technical SEO territory. A team that already handles crawlability and rendering has the skills for it, and mainly needs to know that a new set of bots now depends on them. 10 SEO features a modern CMS should have covers the crawlability side; this article covers the facts those bots need to find once they arrive.
Where else does the answer come from?
Your website is one of several sources a chatbot checks. Community hubs, industry directories, review platforms and ranking articles carry a large share of the citations behind vendor recommendations, because a page somebody else published is easier to trust than your own claim about yourself. G2 found the same pattern from the buyer's side: a citation from a review site lifts confidence in an AI answer more than anything else. A review platform published that finding, so read it with the obvious caveat.
A fact can therefore be missing from an answer even when your own site states it, because nothing outside your site confirms it. Those outside pages then get read twice. Forrester found that buyers validate AI answers against peers, product experts and analysts, so the chatbot reads a directory profile and afterwards the buyer reads it too.
Inconsistency costs double for the same reason. If your capacity, your industries or even your company name appear one way on your site and another way in a profile somebody filled in six years ago, the chatbot holds two versions and no way to choose between them. You need to exist as an entity rather than as a loose string of text: one name, profiles that connect to each other, records a machine can resolve into a single organisation.
The worst version of this is quiet. Your page supplies the answer, a competitor gets named, and the reason is that your content was useful while your identity was ambiguous.
How do you publish all of that without hiring a department?
The requirement is bigger than it was, and it does not end. Prices move, a range changes, an industry gets added, a limitation stops being true. No markup trick substitutes for having the information written down and keeping it right.
The same technology that raised the bar also lowered the cost of clearing it. Four jobs that used to need a team:
- Data straight from the system that holds it. Your prices, dimensions, stock and lead times already live in a PIM or an ERP. An API connection publishes them on the website and keeps them current.
- Specifications and manuals turned into pages. The documents exist already, usually as PDFs nobody opens. A model pulls the values out and drafts the product page, the specification table and the supporting information pages, with an editor checking the numbers before anything goes live.
- What your own people know, turned into text. A twenty-minute conversation with a service engineer becomes an application description. Photos from an installation become a case. That expertise was always in the building. It never got written down because writing it down cost more than it seemed to be worth.
- One catalogue, many languages. Translation opens the same information to buyers in countries you never wrote for, and every language multiplies the questions you can answer.
Applied properly, technology lets a small team do what used to be out of reach: keep a large, detailed website current. Drupal Paragraphs: from unusable to empowering content editors shows how the right content model makes that volume manageable for editors rather than developers.
What is the prize at the end of this work?
Do the work and the marketing maths changes in your favour, particularly for a smaller company. A chatbot checking whether an answer exists has no idea who spent more on marketing.
That inverts twenty years of how attention got handed out. The competitor with the budget you could never match ran the campaigns, took the big stand at the trade fair, sponsored the conference and bought the ads, and for two decades the clicks followed that spending. You were never going to outbid them. You can now answer the question their pages leave open.
Being in the answer beats being everywhere, and it costs less. One accurate specification page can put you in front of a buyer who would otherwise never have come across your name. That is what getting your company recommended by AI looks like in practice: one page with the right number, cited in a conversation you never see.
Buyers already behave this way. One in three in the G2 survey bought from a vendor they had not heard of before the research started, so the shortlist came from a checkable answer rather than from years of brand building. Eighty-five percent said they think more highly of a vendor after a chatbot recommends it, which hands you the credibility a campaign used to pay for.
The buyer also arrives in a better mood. They matched your numbers against their conditions themselves, decided you fit and came to you, which beats any campaign as a way to start a conversation.
What you get is a very warm lead.
How we build this in Drupal for our clients
Everything above asks for hundreds of precise pages, in several languages, with values that stay consistent and markup that matches the content, all of it kept current. That is a content modelling and content operations problem before it is a writing problem, and it is where a page builder runs out of road. It is also most of what we build for clients, so here are the four mechanisms we use and what each one gives you.
We model your specifications as fields instead of prose. Every value gets its own field, so one set of numbers feeds the product page, the comparison table, the structured markup and any outbound feed. You change a number once and every surface follows, which means no forgotten page where last year's price survives. It also means Schema.org markup and metadata get generated from your content rather than maintained by hand beside it.
We generate page families from configuration. Taxonomy and views build and maintain a whole family of pages from one structure. The eleventh industry page then costs a configuration change rather than a project, so enumerating ten industries is one build instead of ten, and the family stays consistent as you extend it.
We put the AI layer inside your editorial flow. Extraction from your specification documents, which starts with choosing the right tool for the documents you actually have. Drafting content and metadata with the AI modules across a large archive. One change reaching hundreds of entities without a developer. Translation with an editor approving each result. All of it happens where your team already edits and publishes, so the volume described above becomes work your own people can absorb instead of a project you outsource every quarter.
We connect the site to the system that already holds your facts. Integrating Drupal with a PIM or an ERP keeps the system of record as the single source, so your website stops disagreeing with it and nobody retypes a specification into a page. The same REST and JSON:API layer hands those values to anything that asks for them in a machine-readable form, so one piece of plumbing serves the buyer reading a page and the software checking it.
Our clients end up with a content system their own team can run. Facts arrive from the system that holds them, the content model keeps them consistent, and their editors stay in charge of what gets published.
The gain is scale for a modest effort. A handful of people keep hundreds of pages of specifications, use cases and industry detail accurate, current and translated, and that is the material a chatbot needs before it will cite a company and recommend it.
Read also: how to keep a multilingual website under control with the right CMS, since every gap described here multiplies by the number of languages you sell in.
This is a good moment to spend your effort on substance. Work that needed a department five years ago now fits a small team, and the loudest campaign no longer wins by default. The company that wins is the one whose real numbers, real limits and real cases are written down where a machine can find them.
Want to get your company recommended by AI in buyer shortlists?
We build content systems for manufacturing and B2B clients where product specifications, industry pages, and structured markup stay accurate across hundreds of pages and several languages. PIM integration, field-based content models, and AI-assisted editorial workflows keep the facts on the site aligned with what buyers ask chatbots to verify.
Interested in making your website quotable by AI answer engines? Our team specializes in Drupal content modelling, enterprise integrations, and editorial operations at scale. Visit our content management solutions to see how we can help you show up when buyers ask for a supplier shortlist.
